[Cryptography] massively parallel processing
phill at hallambaker.com
Thu Sep 15 09:53:33 EDT 2016
On Wed, Sep 14, 2016 at 6:48 PM, John Denker <jsd at av8n.com> wrote:
> On 09/14/2016 10:00 AM, Henry Baker wrote:
> > The multiprocessor revolution is just beginning to take off; I can
> > recall my first computer (IBM 1401) with 4,000 (not 4096!)
> > "characters" of RAM; we're nearly to the point where a consumer can
> > purchase a computer with 4,000 *processors*.
> You can get GPU cards with more than 2000 cores per card. For
> bitcoin mining, or for physics simulations -- or for "consumer"
> applications such as fancy graphics -- you might plug in two or
> more such cards, at which point you're already well over 4000
> processors in a single "computer".
> It's also good for keeping you warm in the winter.
My work machine for over a year now has been a 12 Thread Intel CPU with
twin Nvidia GTX980s with 2048 cores each.
I am not sure quite where it lies on the SIMD/MIMD chart, I know there is
more than one instruction pointer involved but not how many. The latest
versions of DirectX have been gradually making this GPU power available to
programs but I don't think they are there yet.
It might well be time for the industry to start re-learning the things Bill
Dally and David May were talking about in the 1980s when I did my DPhil. on
applying formal methods to a large array of transputers.
What most folk miss about Moore's law is that most of the improvement comes
from more transistors. The transistors themselves have not been getting
much faster for reasons of heat dissipation. And we ran out of useful
things to do to use extra gates to optimize a single core a decade ago.
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the cryptography